While I'd like to keep NA values (aka ineligible feature) separate from genuine FALSE values (tested and did not pass), this causes considerable typing overhead for subsetting the tables by DTU and is an easy and recurrent source of errors in analysis of the results.
Subsetting a table by a vector (logical or keys) always also returns lines that are NA for the criterion. In order to only subset the rows with TRUE values, the criterion must include !is.na(). This behaviour affects all the logical fields, not just DTU.

Based on how many times I've been caught out by this bug/feature of tables/R, I think RATs should do a final pass over the results and replace all NA values in the flag fields with explicit FALSE. Ina sense, items not eligible for testing, automatically are not DTU, so this is not too much of a stretch and it should make downstream analysis safer and easier.